What Matters In On-Policy Reinforcement Learning? A Large-Scale Empirical Study.
Marcin AndrychowiczAnton RaichukPiotr StanczykManu OrsiniSertan GirginRaphaël MarinierLéonard HussenotMatthieu GeistOlivier PietquinMarcin MichalskiSylvain GellyOlivier BachemPublished in: CoRR (2020)
Keyphrases
- empirical studies
- reinforcement learning
- optimal policy
- policy search
- action selection
- markov decision process
- empirical analysis
- real world data sets
- function approximation
- markov decision processes
- reinforcement learning problems
- policy evaluation
- action space
- state space
- approximate dynamic programming
- small scale
- state and action spaces
- reward function
- reinforcement learning algorithms
- partially observable environments
- partially observable
- markov decision problems
- control policy
- function approximators
- machine learning
- temporal difference
- policy gradient
- dynamic programming
- actor critic
- model free
- real world
- decision problems
- average reward
- uci datasets
- rl algorithms
- experimental design
- dynamical systems
- policy iteration
- inverse reinforcement learning
- real life
- learning algorithm
- systematic review
- agent receives
- policy gradient methods
- partially observable domains
- continuous state spaces
- partially observable markov decision processes
- average cost